/Titanic-Machine-Learning-from-Disaster

Made a prediction model from the titanic dataset from kaggle also analyzed the different sizes of NN and the effect of dropout. ---------- model used: NN with dropout

Primary LanguageJupyter Notebook

Titanic-Machine-Learning-from-Disaster

The above folder has the code to analyze the Titanic disaster problem using neural network. This dataset performs better with arguably simpler algorithms like Random forests but since this was the first dataset I ever explored, I wanted to start keras in it.

Different accuracy scores with different NN sizes:

iteration over different NN sizes:

Hidden layer size train_loss train_accuracy dev_loss dev_accuracy dropout Inference
15 10 10 10 0.4592 0.8069 0.4386 0.7982 no overfitting
15 10 0.4171 0.8204 0.4153 0.8161 no overfitting
7 5 5 0.3919 0.8293 0.4239 0.8341 no overfitting
20 5 0.4198 0.8278 0.4271 0.8072 no overfitting
25 5 0.3811 0.8398 0.4522 0.8430 0.1 0 25 best for first layer
30 5 0.3948 0.8308 0.4024 0.8341 0.1 0
25 5 5 0.3657 0.8473 0.4108 0.8430 0.1 0 even better!!!